>>>>> "Peter" == Peter Eisentraut <[EMAIL PROTECTED]> writes:
Peter> Akim Demaille writes:
>> * host or build characteristic?
Peter> It's neither, nor should it matter. It's a feature of the
Peter> compiler. If there's an alternative Cygwin compiler that
Peter> generates .com files, then that's what EXEEXT is. If there's a
Peter> regular Unix compiler that generates a file 'foo.bazoo' when
Peter> invoked 'cc ... -o foo', then EXEEXT='.bazoo'. Also, a
Peter> compiler suite could theoretically choose just about anything
Peter> for OBJEXT, since those files go from the compiler to the
Peter> linker and neither the host nor the build operating system or
Peter> CPU care.
What you describe is precisely what I meant by `build': there is not a
single reference to the host in what you describe. That's also why
just testing the compiler is not what everybody wants. Why? Because
some people might use a cross-compiler which produces foo, not
foo.exe, while the machine on which the executable will be installed
needs to be installed as foo.exe.
Peter> Any solution that uses 'uname', preprocessor symbols, or other
Peter> random features of a particular affected system is not going to
Peter> be general. That's the imake way.
What I described is the rationale for departing from the Autoconf
philosophy: we want to adjust ourselves to something we just can't
check: the host environment.
Why isn't `install' which takes care of this on the *host* machine?